2011 International Conference on Computer Science and Information Technology (ICCSIT 2011)
IPCSIT vol. 51 (2012) © (2012) IACSIT Press, Singapore
DOI: 10.7763/IPCSIT.2012.V51.33
Exploring Cyber Opinion Leaders Based on Social Network Analysis:
Sampling College Student Employment Public Opinion
Zhuozhuo Lia,*, Zihan Dingb
a
School of Society, Soochow University Library,No.199.Ren ai Road,Dushu Lake Campus of Soochow
University,SSIP,SuZhou and 215123, China
b
School of Society, Soochow University,No.199.Ren ai Road,Dushu Lake Campus of Soochow
University,SSIP,SuZhou and 215123, China
Abstract. Combining application research using social network analysis, the opinion leadership could be
identified through the analysis of online public opinion structure. The case study of public opinion about
college student employment is based on the measurement of the degree centrality, core-periphery and
influence coefficient. The online opinion leadership that explored by the combination of social network
analysis and content analysis should be the focus of monitoring public opinion.
Keywords: social network analysis, cyber opinion, opinion leader, college student employment
1. Introduction
As the mirror of reality, cyber opinion is a truthfully reflection of the state of mind among the people; it
delivers state and trend of public opinion of each and every aspects of the society directly and timely. Cyber
opinion platform—internet—is a network of complexity forged with multiple social relation networks in
which one or several core members, also known as cyber opinion leaders, can always be found. Through
their comments, opinion leaders post information, communicate information, bring people into consensus
relatively quickly, brew emotion, and even induce action, alter the opinions of other netizens’, and lead
cyber opinion orientation. Discovering cyber opinion leaders and strengthening the supervision and
management on them is a critical mean of improving cyber opinion management capability. With respect to
the traditional cyber opinion control which identifies, analyzes, and tracks every pieces of information, the
idea of cyber opinion leader discovery is less time-consuming and more efficient. Discovering cyber
opinion leaders in their virtual community using social network analysis, evaluating their influence, and
emphasizing on the analysis and supervision on their comments, can not only control the trend of public
opinion on mass event, but also monitor the spreading of cyber opinion from the source, and even eliminate
negative public opinion information from its source.
2. The Application of Social Network Analysis in the Study Of Cyber Opinion
Leader
2.1. Comments on Methods of Cyber Opinion Study
Currently, the common method used in analyzing the content of cyber opinion is content analysis.
Content analysis is able to describe the certain characterized communicative content objectively,
systematically, and quantitatively. During the research of methodology, Yi Liu [1] discussed mining
internet content, network structure, and internet log using content analysis enables describing the public
opinion information on the internet, inferring the intention, attitude, and emotion leaning of the cyber
* Zhuozhuo Li. Tel.: 15962146201.
E-mail address: smileforever96@126.com.
177
opinion information communication subject, so as to describe and infer the inception and development of
cyber opinion. The application of method mainly concentrated on research of the attitudes and emotions the
public opinions articulate, e.g. Yukiko Kawai and others [2] conducted emotion differential analysis on the
comments of netizens using content analysis on the credibility of on-Column news. Xiaoyun Huang and
others [3] took one step farther to discuss the application of text data mining in content analysis on cyber
opinion, and established cyber opinion information text mining model. Using internet text mining, cyber
opinion text information can be expressed; its characteristic can be extracted, content summarized,
categorized, clustered, relevance and semantics analyzed, and the tendency can be estimated using internet
text, so as to discover focal and controversial topics, and to provide technical support and resolution for
further analysis.
Analysis on cyber opinion can be based not only on the information content of public opinion, but also
on analyzing the cyber structure and the pattern related to public opinion development discovery. Social
network analysis is a method that combines visualization and scalability; it attempts to quantitatively
measure various relationships so as to reveal the structure of group relations and study the effect such a
structure have on the group functionality and individuals in the group. In recent years, social network
analysis is used in the study of cyber opinion. It is especially widely applied in analyzing the relation
structure and group characteristics in virtual community members. Researchers such as Jingmin Chen [4]
introduced social network analysis into the studies on cyber opinion, dissected an internet political group
using internet density, cohesive subgroups, and structural holes three rubrics. They estimated the
transmission pathway of public opinion information according to the degree of structural density of virtual
community, and discovered small groups that dominate the development of public opinion based on the
interrelations between members. These researchers further proposed that a strengthened supervision of the
intentions of the subgroup is needed, in order to help related departments understand cyber opinion, and
consequently make the right decisions.
Concisely, social network analysis concentrates more on the relations between events and participants
compared to other methods. When studying the transmission and relocation of cyber public information,
more emphasis on the pattern of behavioral alternations of netizens that are caused by different cyber
relation structures reveals better the essence of the transmission and relocation of cyber public information
for it puts more emphasis the factor of human.
2.2. The Combination of Social Network Analysis and Cyber Opinion Leader Study
Social network analysis has developed a series of professional theories and models that are applied in
structural analysis of on Column community up until this day. Among these is the famous structural holes
theory often addressed as the classic theory determined to find public opinion leaders. According to Ronald
S. Burt [5], the Author of structural holes theory, when there are only indirect or lack of relations between
people in cyberspace, a hollow “hole” comes into being, which he named “structural hole”. Structural holes
usually make up the “valves” for information reflection and resource liquidity in cyberspace. In cyberspace,
the more “structural holes” one possesses, the more structurally advantageous this actor will be;
subsequently, the higher possibility of return this actor will have through these advantages.
Centrality is another critical indicator in measuring an actor’s status superiority and social prestige.
Alex Bavelas [6] is the first one that conducted researches on the formal characteristics of centrality.
Hereafter, Juhani Nieminen gave this concept a systematic explanation. The so-called point-centrality in
cyberspace is the measure of points; that is, the number of points directly connected to one points. If a point
has a high measure of points, this point is said to be at the center, and the actor this point denotes is the
central figure. This is the easiest and most direct approach to determine a public opinion leader. Besides this,
the theory of “the strength of weak ties” and six degree of separation, both by Mark Granovetter, can also
be applied in the analysis of the status and effect of every member in the cyber community that is under
weak tie status, so as to enrich the cognition of cyber community public opinion leaders.
Some models and methods in social network analysis can be applied in the study on cyber opinion
leaders. The rate of transmission and development of public opinion among the existing virtual circle of the
178
public opinion leaders can be reflected by computing the tightness of social network density. By measuring
centrality and structural holes constraint factor, the public opinion leaders in the circle can be quickly
identified; the position of the leader in the virtual circle can be understood, as well as the influence, so as to
infer the intention and attitude leaning of the cyber opinion leader and seize the opinion trend of the entire
network while track false messages to the origin and erase the negative effect they impose, reinforce the
analysis and estimation of cyber opinion leaders with focuses.
3. The Design of Social Network Analysis Methods in Cyber Opinion Leader
Discovery Study
3.1. Approach
In a virtual community, members connect with each other by replying posts of others’. To any given
member, the sum of replies acquired from other members is denoted as point in degree; the sum of replies
sent to other members is denoted as point out degree. If one member is able to attract the attention of a large
number of members with one post, to be broadly and frequently discussed with a long thread; or one
member is actively replying to posts from other members, this member becomes of the virtual community
an active member. A cyber opinion leader is inevitably an active member of the virtual community. Active
members of a given virtual community can be identified with level of activity explicitly observed by
measuring the point in degree and point out degree. However, active members are not necessarily cyber
opinion leaders. Only those active members who are widely supported and getting mass response are
essentially cyber opinion leaders. For this, analyzing the influence of active members in a virtual
community is the key to locating the cyber leaders in that community. Influence can be expressed using
quantitative measure, determined by calculating the sums of supports and opposes a given member’s
opinion acquired in a given period. In other words, the higher the count of explicit supports this member’s
opinion acquires, the higher the influence; so as to determine the cyber opinion leaders in a virtual
community.
3.2. Factors
The level of activity of members can be easily and directly measured applying centrality analysis and
core-periphery model in social network analysis, so as to determine the position of the member and the
tightness between active members. Denote “coefficient of influence” as category index to filter virtual
community cyber opinion leaders from active members.



Centrality：Centrality is the quantitative indicator in measuring the power of an actor, the purpose of which is to
locate the core members in a network in the entire network structure. Point degree centrality is an important
method in measure of centrality; it measures the capability of a give point develops connection with other points.
Suppose one actor has direct connection with multiple other actors, this actor is at a central position, and thus
holds relatively high power.
Core-periphery Model：Core-periphery model is a quantitative of the actor’s position. This model identifies
series of actors with higher density (core) and lower density (periphery) to determine the positions of actors in a
network based on the degree of dispersion of the connection between members.
Coefficient of Influence：Coefficient of influence refers to a given member’s degree of influence on other
members in a virtual community. The higher the influence in the virtual community, the higher degree of approval
this member has. To measure the influence of a given member, value of response and value of approval need to be
measured. Value of response is the sums of both supports and opposes a member acquire in a given period. Value
of approval is degree of approval this member acquired from other members, a. k. a. total support count.
Coefficient of influence is the ratio of value of approval over value of response.
3.3. Executive Approach
To discover cyber opinion leaders, a virtual community should first be selected, related logs of
discussion participants be extracted. We create an interaction matrix, assuming that a connection is
generated by users replying posts of each other, that is, if user X replied user Y’s post, X and Y has a
connection, regardless of the reply one user has for another is one time or multiple times, denoted as “1”;
otherwise, if there is no connection between X and Y, denoted as “0”. By measuring the centrality of users,
locate the users at the central position, a. k. a. active members. The position of active members can be
179
determined using the CORR algorithm in core-periphery model. Lastly, compute the coefficient of
influence of active members individually, and the one with the largest coefficient of influence is determined
as the opinion leader of this virtual community.
4. College Students Employment Public Opinion Case Analysis
College student employment is one of the topics netizens constantly follow. Using cyber opinion, the
psychology and attitude of how the netizens treat the current state of College student employment can be
understood, which is beneficial for the government and universities to give targeted guidance to college
students looking for jobs. Extract logs of every post and reply before January 28, 2011 from Tianya forum
college student employment module, and apply social network analysis method to discover the cyber
opinion leader, whose influence will be inspected.
4.1. Social Network Interaction Matrix of Virtual Community and Charts
In this virtual community, connection between members is created by replying to each other’s posts.
After a period of collecting, the interactions between the 47 members of this virtual community are made
into a matrix, denoting members alphabetically from “1” to “47”. Column denotes replier, row denotes
poster. A member replies another is denoted as “1”; otherwise, they have no connection, is denoted as “0”.
Thus, a 47 by 47 matrix is formed, which is able to explicitly depict the network relation in this virtual
community.
Fig. 1.The Network Relations between 47 Members in the Virtual Community
It’s not difficult to see that 13, 16, and 19 are at the central position of the network, having many other
points connected to these points. They are active members in the community, having a high count of
followers in this virtual community. Members like 42 is at the peripheral position, having hardly any direct
connections to other points, which indicates that they barely communicate with other members in the
community, seldom express their opinions in the community and make little contribution.
4.2. Measuring Centrality
The discovery of the cyber opinion leaders in a virtual community can be achieved by measuring the
point degree centrality of that community. Input the interaction matrix into Ucinet to compute the centrality
of the matrix, shown as table 1:
180
Table 1. Centrality measure result (partial)
OutDegree
InDegree
NrmOutDeg
NrmInDeg
16
8.000
19.000
8.696
20.652
35
6.000
2.000
6.522
2.174
37
4.000
0.000
4.348
0.000
26
3.000
1.000
3.261
1.087
39
3.000
2.000
3.261
2.174
19
3.000
26.000
3.261
28.261
38
3.000
3.000
3.261
3.261
31
3.000
2.000
3.261
2.174
9
3.000
1.000
3.261
1.087
46
3.000
0.000
3.261
0.000
42
2.000
0.000
2.174
0.000
2
2.000
1.000
2.174
1.087
45
2.000
2.000
2.174
2.174
14
2.000
2.000
2.174
2.174
13
2.000
11.000
2.174
11.957
From the analysis, each member is represented by different point in degree and point out degree.
According to the difference in point degree centrality, the power and statue of the members in the
community vary. The analysis shows that member 16 and member 19 are ranked top in both point in degree
and point out degree compared to the residual members, which indicates that they are relatively active in
the virtual community. Not only do their posts attract a large number of other members, they also positively
engaged in the discussion of others, making high contribution to the virtual community. They are the active
members in the community. The points in degrees are larger than points out degrees, indicating that their
comments are much more followed by the residual members. Member 35 has a point out degree far larger
than the point in degree, indicating active following on other members, yet was less followed by others.
Member 13 has a point in degree far larger than point out degree, indicating certain influence the comments
this member have, followed by many members, but seldom does this member follow comments of others.
4.3. Positions of Core Members in Core-Periphery Model
Identify the position of each member based on the computation from the CORR algorithm in coreperiphery model using the software Ucinet. Arrange the result into table 2. The members who have a core
index of larger than 0.15 form the core population; the ones have between 0.08 and 0.15 form semiperipheral population, and the residual form peripheral population.
Table 2. Centrality and position of 47 members
Position
core
Member
Centrality
Member
Centrality
35
0.471
28
0.116
16
0.345
47
0.116
37
0.309
22
0.116
38
0.227
4
0.099
19
0.209
14
0.083
39
0.172
6
0.054
42
0.163
30
0.054
13
0.163
32
0.054
2
0.163
20
0.054
25
0.163
5
0.054
Position
Semiperiphery
Periphery
181
Semiperiphery
17
0.163
40
0.054
31
0.156
15
0.054
9
0.153
18
0.044
26
0.142
8
0.044
46
0.139
33
0.044
41
0.132
43
0.044
21
0.116
11
0.044
1
0.116
3
0.044
45
0.116
10
0.044
36
0.116
24
0.044
23
0.116
29
0.044
7
0.116
44
0
34
0.116
27
0
12
0.116
In table 2, member 13, 16, and 19 appear in the core population again, while the residual members falls
into the peripheral position. This further verified the central position of the three members’. Meanwhile,
this result also indicated these active members capture the crucial position in the information pathway in the
virtual community.
4.4. Measuring the Influence of Cyber Opinion Leaders
From the above analysis, member 13, 16, and 19 three people can be explicitly seen as the core
members of the virtual community. Their posted opinions are responded by the super majority of the
members, have a high degree of attraction. However, the responses are mixed. Whether their opinions are
approved by the majority of the members, causing resonances, imposing influence on other members, are
significantly important in judging the trend of cyber opinion. To speculate whether active members are able
to influence the acts of other members, every opinion in the virtual community that involves cyber opinion
leaders and the residual members needs to be sampled, and computed by coefficient of Influence, to
ultimately discover the real cyber opinion leaders. The statistics is shown in table 3.
Table 3. Coefficient of influence of active members in virtual community
Member
Value of Response
Value of Approval
Coefficient of Influence
13
74
27
0.36
16
102
84
0.82
19
118
86
0.73
In table 3, coefficient of Influence reflex on the degree of influence active members have on the virtual
community. Empirical evidence suggests despite the fact that all of the three members are able to gather
huge popularities, their recognitions acquired from others come in significant differences. Obviously,
member 16 and 19 both has high values of response and values of approval, as well as a relatively high
coefficient of Influence, which reflect that these two members are the real cyber opinion leaders of this
virtual community. They enjoy relative high prestige in the virtual community, and their comments resonate
with netizens, even causing mass response, and guide the developmental trend of cyber opinion. The other
member, 13, is otherwise: the value of oppose is far higher than value of approval, which is to say that the
comments by member 13 are less often approved by the population, are unable to impose influence on
182
others’ opinions, and even at times objected by others, having a lesser effect on the accumulation of cyber
public opinion.
We randomly withdraw a log containing a cyber opinion leader’s thread. In it, the cyber opinion leader
16 in Jan 18, 2011 expressed contempt against the behavior posted in a news alleging college students
“would rather stay in metropolis as an ant, than go back to town as a noble”, saying: “the cost of living and
competition are low in towns. It’s better to be a head of a dog than to be a tail of a lion. A person has to live
with dignity”. This comment received 20 relies in total, with 13 supports. Meanwhile, based on observation,
member 16 receives replies for every post from certain members, forming a subgroup with steady relations.
Not only do they respond to the opinion, they also actively resonate. Therefore, a group tightly unified
around the cyber opinion leader as its center is formed.
5. Conclusion
The application of social network analysis has helped discover the notable variance in terms of
influence each users in a virtual community has according to the difference of their characters and network
positions. The measuring of centrality depicts that the ones who score high centrality and are at central
positions are active members in the given social network. Comments from these members are broadly
followed by the majority of the residual members, while inflecting great promotional effect on the
propagation of public opinion. Yet to identify whether they are opinion leader or not also requires taking
the influence of the members in the given community into account. The comments from active members
stand out from the extensive and heterogeneous flow of cyber information, in certain levels guide the trend
of cyber opinion and even catalyze the development of events only when they are broadly supported by
other members. In most cases, cyber opinion leaders will process the pieces of information they acquired
before they spread and communicate them to a broader audience in form of messages. These messages
which contain the objective opinions from the cyber opinion leaders are able to tweak or dominate the
attitudes and behaviors of the followers and the influenced. The research has revealed that the social
network analysis discovers active members in a community simply using the interaction between actors in a
network; however, to become public opinion leaders requires not only the initiatives in interaction, but also
relatively high prestige among members. Therefore, when discovering cyber opinion leaders, assistance
from corresponding content analysis is much needed, calculating also the emotion, proposal, and attitude in
the explicit opinions to ensure the accuracy of cyber opinion leader discovery.
Easily drawing conclusion based on the above analysis, comments from cyber opinion leaders shall
become focal subjects under cyber opinion supervision and management compared to ones from the
residual members. Against sensitive and special topics the society is concerned about, identify the cyber
opinion leaders, specially supervise, control and timely handle their inflammatory comments to avoid
negative emotions to continually affect the residual members. Based on that, perfect the early warning and
crisis responding mechanisms against cyber opinion leaders. In the daily internet management, cyber
opinion leaders can be purposely fostered. The prestige they have in the virtual community can be used to
guide public opinion positively, and guide people to fight back false and reactionary speeches actively.
6. References
[1] Yi liu. Application of Content Analysis Method in Analysis of Net Mediated Public Sentiment. Journal of Tianjin
University(Social Sciences) , 2006,28(4): 307-310 (in Chinese).
[2] Jianwei Zhang, Yukiko Kawai, et al. Sentiment Bias Detection in Support of News Credibility Judgment. HICSS
2011 Minitrack: Web Information Credibility Analysis Forty-fourth Annual Hawaii International Conference on
System Sciences, 2011.
[3] Xiaobin Huang, Chao Zhao. Application of Text Mining Technology in Analysis of Net-Mediated Public
Sentiment. Information Science, 2009,27(1): 94-99 (in Chinese).
[4] Jingmin Chen, Yongzhuan Han. The Analysis of Internet Public Opinion Based on Mining of Virtual Network.
Manufacture Information Engineering, 2010,39(5) (in Chinese).
183
[5] Burt R. Structural Holes:The Social Structure of Competition. Cambridge: Havard University Press,1992.
[6] John Scott, Jun Liu. Social Network Analysis:A Handbook. Chongqin:Chongqin University Publication,2007.
[7] Ran Chen. The Exploration on the BBS Opinion Leadership Measuring Methods.
http://www.tsjc.tsinghua.edu.cn/newmedia/wordpdf/10. pdf(in Chinese).
[8] Ming Zhou, Jingwei Zhuang. The Application of Social Network Analysis in Management of BBS. Technology
Economics, 2009, 28(11): 93-98 (in Chinese).
[9] Hong Yu. The Exploration on the BBS Opinion Leadership Filtering Model. Journalism &
Communication ,2008, 15(2): 66-75 (in Chinese).
184